Files
vigilcare-clinical/docs/clinical-testing-guide.md
T

16 KiB
Raw Blame History

VigilCare Clinical Testing Guide

For doctors and nurses evaluating the alerting dashboard

This guide walks you through testing the VigilCare monitoring dashboard. You will review simulated patient scenarios, decide whether alerts are clinically meaningful, and record structured feedback that helps the team improve the system.

No programming knowledge is required. A facilitator will start the backend services and open the dashboard in your browser.


Table of Contents

  1. What you are testing
  2. Important notes before you start
  3. Getting to the dashboard
  4. Tour of the application
  5. Your core task — review and rate alerts
  6. Understanding the six feedback ratings
  7. Recommended testing sessions
  8. Scenario scripts
  9. Session checklist
  10. Submitting your feedback
  11. Frequently asked questions

1. What you are testing

VigilCare is a clinical decision support prototype. It watches vital signs and lab-style observations for hospitalized patients and raises alerts when something looks wrong — for example:

  • A single vital sign in a warning range (heart rate, blood pressure, temperature)
  • A NEWS2 early warning score crossing medium or high risk
  • SIRS / sepsis or qSOFA patterns suggesting infection or organ dysfunction
  • Rapid deterioration — a vital changing quickly even if still “in range”
  • Sepsis bundle tracking — whether time-critical treatments were ordered and completed

The dashboard is where you, as a clinician, would see those alerts and decide what to do. This testing round adds structured feedback: after reviewing each alert, you tell us whether it would help or hinder real clinical work.

Your ratings are the primary output of this study. There are no right or wrong answers — we want your honest clinical judgment on simulated cases.


2. Important notes before you start

Simulated patients only

All patients in this test are fictional. Data is generated by a replay simulator, not from real bedside monitors. Treat it like a training exercise, not live clinical work.

Not a production EHR

This dashboard is a research and evaluation tool. It does not replace your hospitals charting system, does not send real pages, and does not write orders to a live pharmacy or lab.

Acknowledge and resolve are practice actions

You can tap Ack and Resolve on alerts to walk through the workflow. In this environment those actions update the test database only.

Feedback stays in your browser (for now)

Your ratings are saved in the browser on the computer you use. Export them at the end of the session (see §10) and send the file to the study facilitator. If you switch computers or clear browser data without exporting, ratings may be lost.

Default clinician ID

Acknowledging alerts uses a demo identifier (DR-DEMO) unless the facilitator configures yours. This does not affect your feedback ratings.


3. Getting to the dashboard

The facilitator will ensure the following are running. You only need the browser URL.

What Where
Dashboard http://localhost:5173 (or URL provided by facilitator)
Your role Review alerts and submit feedback

On first open you land on Virtual Ward — a list of active simulated patients sorted by acuity (NEWS2 score).

If the ward list is empty, ask the facilitator to start or replay a scenario (see §8).


4. Tour of the application

Use the sidebar (desktop) or bottom navigation (mobile) to move between screens.

Virtual Ward

Purpose: See who is on the floor and who needs attention first.

  • Patients sorted by NEWS2 score (higher = higher concern on this board).
  • Badge shows count of patients with NEWS2 ≥ 7.
  • Filter by department if asked (ICU, General Medicine, Surgery).
  • Click a patient row to open their detail page.

What to notice: Does the sort order match how you would prioritize a real ward round?


Patient Detail

Purpose: Deep review of one patient — vitals, scores, alerts, trends, and why alerts fired.

Section What it shows
Scores Current NEWS2 total and risk level
Latest Vitals Most recent heart rate, RR, BP, SpO₂, temperature, etc.
Active Alerts Open alerts for this patient — click a row to see reasoning
Alert reasoning Plain-language explanation of why the alert fired; may show recent medications
Orders Clinical orders (labs, antibiotics, fluids, etc.)
Sepsis bundle If sepsis was suspected — four time-critical elements and compliance status
Vital sign charts Trends for HR, RR, systolic BP, SpO₂, temperature
NEWS2 history How the early warning score changed over time
Replay controls Local timeline bar (pause, speed, Next Alert →) to step through the case

What to notice: Would you trust these charts and explanations during a real handoff? Is anything missing?


Alert Center

Purpose: Hospital-wide inbox — all alerts across patients, not just one encounter.

  • Tabs: Open, Acknowledged, Resolved, Escalated.
  • Each card shows severity, type, details, and time.
  • Ack / Resolve for workflow practice.
  • Feedback buttons on every card (see §5).

What to notice: Is it easy to triage multiple patients from one screen? Would you use this during a shift?


Feedback Summary

Purpose: Aggregate view of all ratings you (and others on the same browser) have submitted.

  • Total ratings, % useful / would act, % false positive, timing issues.
  • Breakdown by alert type (e.g. sepsis, NEWS2, heart rate warning).
  • Recent feedback list with your notes.
  • Export JSON and Export CSV — use CSV for spreadsheets.

Open this at the end of a session to sanity-check your work before export.


5. Your core task — review and rate alerts

For each alert you review, complete this short workflow:

1. Read the alert (severity, type, details, time)
2. Optional: open Patient Detail → click the alert → read "why it fired"
3. Optional: look at vitals, charts, medications, sepsis bundle
4. Choose ONE feedback rating (required for that alert)
5. Optional: tap "+ Note" and add a short clinical comment
6. Move to the next alert

Where feedback buttons appear

  • Alert Center — below each alert card.
  • Patient Detail — below the alert reasoning panel (after you click an alert).
  • Active Alerts list — on expanded reasoning view when integrated.

You can rate alerts in any status (Open, Acknowledged, Resolved, Escalated). You may change your rating by selecting a different button — the latest choice replaces the previous one.

Tips for consistent ratings

  • Rate based on what you knew at the time the alert fired, using the charts and reasoning as context — not hindsight after the full scenario finished.
  • If an alert was technically correct but not actionable, consider Too early, Too late, or Missing context rather than Useful.
  • If the patient was stable and the alert reflected expected treatment (e.g. beta-blocker bradycardia), False positive or Missing context may fit better than Useful.
  • Use Would act when you would genuinely change management based on that notification alone.

6. Understanding the six feedback ratings

Button When to use it Example
Useful Clinically appropriate alert; right concern at right time NEWS2 rises to 6 as RR and HR worsen in sepsis scenario
Would act You would change assessment, monitoring, or treatment because of this alert qSOFA alert on a patient you would escalate to senior review
Too early Directionally right but fired before you would act Warning HR while patient still asymptomatic and trending stable
Too late Real problem existed but alert came after you would have intervened Deterioration alert after you would already have called a rapid response
False positive Alert should not have fired for this patient context Bradycardia warning on a patient on scheduled metoprolol with baseline low HR
Missing context Alert may be valid but message lacks information you need HR warning without noting recent beta-blocker dose

Optional notes — what to write

Short phrases are enough. Examples:

  • "Expected bradycardia — patient on metoprolol 25 mg PO"
  • "Would have acted on NEWS2 ≥ 7 sooner if bundle status visible"
  • "Useful but duplicate of HR warning 10 min earlier"
  • "Needed lactate trend, not just single value"

Session A — Quick orientation (2030 minutes)

Goal: Learn the UI and rate at least 5 alerts.

  1. Facilitator replays stable-baseline-01 — confirm ward stays quiet or low acuity.
  2. Facilitator replays uti-sepsis-elderly-01 at faster speed.
  3. You: Virtual Ward → open patient → review 23 alerts with reasoning + charts.
  4. You: Alert Center → rate remaining alerts.
  5. Feedback Summary → export CSV.

Session B — Alert quality deep dive (4560 minutes)

Goal: Compare alert types across scenarios.

  1. medication-false-alarm-01 — focus on false positives and missing context.
  2. uti-sepsis-elderly-01 — sepsis, NEWS2, bundle panel.
  3. respiratory-failure-asthma-01 or post-op-hemorrhage-01 — deterioration patterns.
  4. Export CSV with notes on at least 10 alerts.

Session C — Ward workflow (30 minutes)

Goal: Test prioritization and handoff usability.

  1. Facilitator runs replay-all on the scenario folder (or 23 scenarios back-to-back).
  2. You: Stay on Virtual Ward — note sort order as new patients appear.
  3. Round on each high-NEWS2 patient — detail page only, no Alert Center until end.
  4. Document: Would this order match your morning ward round?

8. Scenario scripts

The facilitator runs these from a separate terminal. You watch the dashboard update.

Scenario Clinical story What to evaluate
stable-baseline-01.json Stable inpatient, minimal abnormal vitals Alert noise — should few or no alerts fire?
medication-false-alarm-01.json Beta-blocker, baseline low HR, boundary vitals False alarms, alert fatigue, medication context in reasoning
uti-sepsis-elderly-01.json Elderly UTI progressing to sepsis SIRS/qSOFA, NEWS2 trend, sepsis bundle, usefulness of escalation
respiratory-failure-asthma-01.json Asthma exacerbation RR/SpO₂ warnings, NEWS2, rapid deterioration
post-op-hemorrhage-01.json Post-operative bleeding BP/HR trends, critical vs warning timing
cardiac-arrest-post-mi-01.json Post-MI deterioration High-acuity alerts, would-act vs too-late
dka-electrolyte-01.json Metabolic emergency Multi-parameter scoring, order visibility
hypothermia-elderly-01.json Temperature-driven risk Temp warnings, NEWS2 contribution

Facilitator command (example):

dotnet run --project VigilCare.Simulator -- replay \
  VigilCare.Simulator/Scenarios/List/uti-sepsis-elderly-01.json \
  --speed 60

--speed 60 means 60× real time — a 3-hour scenario completes in a few minutes. Ask the facilitator to pause between scenarios if you need more review time.


9. Session checklist

Use this during or after your session.

Virtual Ward

  • Patient list loads and refreshes
  • NEWS2 sort order feels clinically sensible
  • Department filter works (if tested)
  • Patient detail opens from row click

Patient Detail

  • Latest vitals match what you expect for the scenario
  • NEWS2 score and risk level display correctly
  • Clicking an alert shows reasoning panel
  • Medication context appears when relevant (beta-blocker scenario)
  • Vital sign charts show sensible trends
  • NEWS2 history chart updates over time
  • Sepsis bundle panel appears when sepsis alerts fire
  • Next Alert → scrolls to review section and selects alerts

Alert Center

  • Open / Acknowledged / Resolved tabs filter correctly
  • Acknowledge and resolve workflow understandable
  • Feedback buttons visible on every card

Feedback (Phase 19)

  • Rated at least 5 alerts (more is better)
  • Used more than one rating category (not all “Useful”)
  • Added notes on at least 2 alerts where context mattered
  • Changed a rating intentionally — confirm new selection sticks after refresh
  • Feedback Summary totals look correct
  • Exported CSV (and JSON if requested)

Overall clinical judgment

  • Which alert types were most trustworthy?
  • Which caused unnecessary noise?
  • What information was missing from alert text or reasoning?
  • Would you want this on a real ward? Why or why not?

10. Submitting your feedback

At the end of your session:

  1. Open Feedback Summary (sidebar → Feedback).
  2. Review totals and recent entries.
  3. Click Export CSV.
  4. Save the file (e.g. vigilcare-feedback-dr-smith-2026-06-20.csv).
  5. Send to the facilitator via the method they specify (email, shared drive, study portal).

The CSV contains: alert ID, alert type, severity, your rating, notes, and timestamp. The research team aggregates exports from all participants.

Optional: Export JSON if the facilitator requests machine-readable format.

Same computer: Ratings persist if you refresh the page on the same browser. Different computer: Export before switching devices.


11. Frequently asked questions

I dont see any patients.
The simulator may not have run yet, or all encounters are discharged. Ask the facilitator to replay a scenario.

Charts are empty but vitals show data.
Wait a few seconds — the page polls every 5 seconds. If charts stay empty, tell the facilitator.

Can I rate the same alert twice?
You can change your rating; only the latest is kept per alert.

Do I have to acknowledge before rating?
No. Rate any alert in any status.

Whats the difference between Useful and Would act?
Useful = good alert clinically. Would act = you would specifically change care because of it. An alert can be useful information but not change your plan — use the button that best matches your reasoning.

The replay bar doesnt pause the simulator.
Correct — replay controls only scrub data already loaded in the browser. The facilitator controls simulator speed separately.

Dark mode?
Toggle in the header if your eyes prefer it; all screens support dark mode.

Who do I contact with problems?
Speak to your session facilitator. Technical issues (blank screen, errors) may need them to restart the API or dashboard.


Quick reference card

I want to… Go to…
See all patients by acuity Virtual Ward
Deep-dive one patient Click patient → Patient Detail
Triage all hospital alerts Alert Center
Understand why an alert fired Patient Detail → click alert → Reasoning
Rate an alert Feedback buttons under alert card or reasoning
See my ratings aggregate Feedback Summary
Submit results Feedback Summary → Export CSV

Thank you for participating. Your clinical feedback directly shapes whether VigilCare alerts help or harm real ward workflows.